Best practice for this workflow I am beginner

[Umar_Farooq]

this workflow trigger from google chat space user prompt in the chat to fetch and download file from the drive nodes required ID’S … I return ID’S by given prompt to the agent what is the best way that I receives ID’S as input of every node instead of by giving system prompt to return id and I also want LLM that understand my space query and run tools according to it … I want to expand this workflow like email and calender tell me best practice to coordinates node input and output

prompt

You are a digital assistant integrated with Google services. Your main task is to operate different Google Drive and Chat operations using the available tools.

Tools
You have access to the following tools:

1. Search_file – Search Google Drive for a file.

   • Input: file or folder name
   • Output: Must return the file name and file ID in JSON format.

2. Download_file – Download the file using the file ID.

   • Input: file ID
   • Output: Must return a download ID.

3. Response_Chat – Send a message back to Google Chat.

Instructions
When a webhook is triggered, the user’s message will indicate which file they are looking for.

1. Use Search_file to look for the file in Google Drive.

2. If a file is found:

   • Extract the file ID from the Search_file result.
   • Use Download_file with that file ID to get the download ID.
   • Then use Response_Chat to respond with a message like:
     “Your file has been found and downloaded successfully. Download ID: [download_id]”

3. If no file is found:

   • Use Response_Chat to respond with:
     “Sorry, the file you’re looking for was not found. Please check the name and try again.”
     My workflow
     
     

[Erick_Torres]

From my point of view, these could be best practices for your flow:

1. The AI ​​Agent node can invoke tools like Search_file and Download_file. The important thing is that the prompt is well-structured so the model understands the flow.

2. Your prompt should contain something like:

3. You are an assistant integrated with Google services. When a user sends a file name, use “Search_file” to look for it. If found, use the resulting “fileId” to call “Download_file”. Respond back with the download ID using “Response_Chat”. If not found, notify the user accordingly.

4. Avoid passing IDs manually; this can be achieved by dynamically connecting the output of Search_file to the input of Download_file using the AI ​​Agent or an intermediate Code node. Even better, structure the output of Search_file so that the fileId is clearly defined as a JSON field that the model can retrieve.

5. If you’re using AI Agent, make sure the Search_file output contains the fileId as a clear key ({ “fileId”: “xyz”, “fileName”: “abc.pdf” }).

6. The prompt should make it clear that this ID should be passed to the Download_file node.

7. If you have a clear agent output structure, you should define it in the prompt in the agent’s output parser.